The invention relates to an information processing system constructed by a memory device and a memory control apparatus having a plurality of cache memories and, more particularly, to a memory control apparatus of an information processing system in which a memory device uses a technique of RAID.
In a conventional memory control apparatus having a cache memory, in response to a write instruction from a central processing unit, an end of writing process is reported to the central processing unit at a time point when write data has been stored into the cache memory in the memory control apparatus. The storage of the write data into a disk device is executed by the memory control apparatus asynchronously with the write instruction from the central processing unit. Such a control is called a write-after control. In the write-after control, when the end of writing process is reported to the central processing unit, since data is not transferred to the disk device, a high-speed response to the central processing unit can be performed.
In the data stored in the cache memory, the data before it is transferred to the disk device is called dirty data. On the other hand, in the data stored in the cache memory, the data which coincides with the data in the disk device is called clean data. In an information processing apparatus for performing the write-after control, the dirty data is data which exists only on the cache memory in the memory control apparatus. Therefore, the dirty data cannot be deleted until the transfer to the disk device is finished.
A data transfer speed between the cache memory and the disk device is slower than a data transfer speed between the central processing unit and the cache memory. There is, consequently, a case where the cache memory is filled with the dirty data at the time of the write-after control. Therefore, a memory control apparatus disclosed in JP-A-4-264940 monitors an amount of dirty data in the cache, an inflow amount of write data to the cache memory of each memory device, and an outflow amount of dirty data due to the write-after process. When the dirty data amount in the cache reaches a certain threshold value (permission amount), a limitation of the allocation of the cache memory to the write data and a data writing preferential execution to the memory device are executed for the memory device in which (the inflow amount) > (the outflow amount). Owing to the above control, a deterioration in performance of the whole system that is caused since the cache memory enters a full state is avoided and the distribution of the allocation of the cache memory for every memory device can be properly performed.
In the write-after process, since the cache memory is volatile, there is a possibility such that the dirty data is extinguished due to a fault of a power source of a control apparatus or the like, a fault of a hardware of the cache memory, or the like. Therefore, in "Evolution of the DASD storage control" disclosed in IBM SYSTEMS JOURNAL, Vol. 28, No. 2, 1989, a memory control apparatus has therein a cache memory and a non-volatile memory (NVS) of a small capacity for backup of dirty data and the dirty data is duplexed and held in the cache memory and non-volatile memory (NVS), thereby preventing an extinction of the dirty data.
Further, a memory control apparatus having a plurality of cache memories which are backed up by a battery is disclosed in JP-A-5-189314. The memory control apparatus has a cache memory whose power source is duplexed by a battery backup. The cache memory is divided into N portions and data read out from a memory device by a command from an upper apparatus is stored into an arbitrary one of the N divided portions of the cache memory. Data which is written into the cache memory by an instruction from the upper apparatus is stored into arbitrary two of the N divided portions of the cache memory. Thus, a guarantee performance of the data at the time of a fault is improved and a deterioration of the performance can be prevented.
In case of using the technique of RAID for the disk device as a memory device, the memory control apparatus needs to form parity data to guarantee data at the time of the write-after process. Therefore, not only the dirty data (write data) but also the clean data (data which is updated by the write data) need to be held on the cache memory.
In case of the memory control apparatus disclosed in JP-A-5-189314 in which data is duplexed and stored into a plurality of cache memories, therefore, there is a case where a balance of an amount of data which cannot be deleted until the data transfer to the disk device is completed is largely broken between the duplexed cache memories in depending on an allocating method of the cache memory for the clean data. When an amount of dirty data (including the clean data to form parity data) in one cache memory reaches a cache memory capacity or a preset dirty data threshold value (permission amount), even in the case where there is a usable area in the other cache memory, the dirty data cannot be duplexed. Thus, a response by the write-after process cannot be performed and the performance deteriorates.
Fig. 8 shows a case where a cache surface (cache to store clean data and dirty data as updated data of the clean data) and an NVS surface (cache to store the dirty data) of each of a cache A 80 and a cache B 81 of a memory control apparatus are fixedly used.
In case of using one of the two cache memories as a cache surface and the other as an NVS surface, as shown in Fig. 8(a), an amount of dirty data in the cache A 80 that is used as a cache surface is larger than an amount of dirty data in the cache B 81 that is used as an NVS surface. Thus, as shown in Fig. 8(b), when the cache A 80 which is used as a cache surface is filled with the dirty data, in spite of the fact that a new usable area c still remains in the cache B 81, the dirty data cannot be duplexed. Therefore, the write-after process cannot be performed.
A memory control apparatus with similar disadvantages as the above-discussed prior art is disclosed in Jai Menon et al: "The Architecture of a Fault-Tolerant Cached RAID Controller"; Computer Architecture News, Vol. 21, No. 2, May 1993, pages 76 to 86, from which the first portion of claim 1 starts out. The.apparatus disclosed in this document has two cache memories and a cache memory control unit for controlling the use of the cache memories during data transfer between a central processor and a memory device both connected to the memory control apparatus.
Updated data from the central processor is received by the memory control apparatus and stored in the two cache memories in duplication. The original data to be updated is read from the memory device always into a predetermined one of the cache memories (if not already there) for use together with the updated data in obtaining updated parity information.
Hence, the updated data (dirty data) are written in duplication into both cache memories for safety against failure of one of them. And the data to be updated and read from the memory device (clean data) is additionally written always into one predetermined cache memory which will thus overflow easily while a large portion of the other remains unused.
It is therefore an object of the invention to provide a memory control apparatus and method for improved efficiency in using cache memory space.
This object is solved by the apparatus of claim 1 and the method of claim 5. The subclaims are directed to preferred embodiments of the invention.
A memory control apparatus according to one embodiment of the invention comprises: a channel control unit, interposed between a central processing unit and a memory device for storing data, for controlling a data transfer to/from the central processing unit; a drive control unit to control a data transfer to/from the memory device; a plurality of cache memories for temporarily storing data that is transferred between the central processing unit and the memory device; and a cache memory control unit having selecting means for selecting the cache memory to store data which is transferred from the memory means. The selecting means arbitrarily selects a first cache memory which is used as a cache surface and a second cache memory which is used as an NVS surface from among a plurality of cache memories. There is a case where the cache memory control unit of the memory control apparatus has memory means for storing control information of the memory control apparatus.
When the data is read out from the memory device, the memory control apparatus transfers the read data to the central processing unit and stores into the cache memory. In this instance, the memory control apparatus arbitrarily selects the cache memory which is used as a cache surface from a plurality of cache memories and stores the data into the selected cache memory.
When the data is written into the memory device, on the other hand, the memory control apparatus stores the write data transferred from the central processing unit into the cache memory. In this instance, when the data which is updated by the write data has been stored as clean data in the cache memory, the cache memory is used as a cache surface and clean data and dirty data as updated data for the clean data are stored. The memory control apparatus arbitrarily selects the cache memory which is used as an NVS surface from a plurality of cache memories and stores the dirty data into the cache memory.
As a specific method when the memory control apparatus selects the cache memory, there is a method of selecting the cache memory on the basis of a virtual or physical address or an identifier of data such as a sequence number or the like in case of dividing into management units.
As another embodiment, there are provided: means for storing usable/unusable state information of each of the unit memories constructing the cache memory into the control memory of the memory control apparatus; and means for calculating the number of usable unit memories of each of the cache memories with reference to the usable/unusable state information of each of the unit memories, and wherein the cache memory is selected by the usable unit memory.
In another embodiment there are provided: means for storing memory use amount information of each of the cache memories into the control memory of the memory control apparatus; means for arbitrarily selecting the first cache memory and the second cache memory from a plurality of cache memories on the basis of the memory use amount information of each of the cache memories; and means for storing dirty data amount information of each of the cache memories or clean data amount information to form parity data into the control memory, and wherein the cache memory is selected on the basis of the dirty data amount information of each of the cache memories or the clean data amount information to form the parity data.
When the data is read out from the memory device, the memory control apparatus selects the cache memory which is used as a cache surface from a plurality of cache memories and stores the read data as clean data into the selected cache memory. In case of updating the clean data, the memory control apparatus selects the cache memory which is used as an NVS surface from the plurality of cache memories and stores the write data into each of the cache memories which are used as a cache surface and an NVS surface. In this instance, the clean data stored in the cache memory that is used as a cache surface is also held on the cache memory.
When, as mentioned above, the memory control apparatus arbitrarily selects the cache memories to be used as a cache surface and an NVS surface from the plurality of cache memories, a deviation of the using states among the plurality of cache memories can be reduced.
In the case where the memory control apparatus selects the cache memory on the basis of an address of the data or a sequence number of the data, the data of a certain group unit having certain characteristics is divided and allocated into the plurality of cache memories by the address or sequence number of the data. Therefore, expect values of probabilities at which the data in the cache memory is updated are averaged among the plurality of cache memories and a cache memory space can be effectively used. That is, the probability of the data to be updated differs depending on the characteristics which the data has. Therefore, by allocating a group of data to the same cache memory, the occurrence of a deviation of the dirty data amounts among the plurality of cache memories in accordance with the data characteristics is prevented, so that which cache memory is allocated to the divided data is decided by an identifier of the data.
In the case where the memory control apparatus selects the cache memory on the basis of the number of requesting times of the allocation of the cache memory, the memory control apparatus has means for holding the number of times of the cache memory selecting process or the number of blocks allocated onto the control memory. The memory control apparatus selects the cache memory on the basis of the number of processing times or the number of blocks. Therefore, since the expect values of the probabilities at which the data in the cache memory is updated are averaged among the plurality of cache memories, the cache memory space can be effectively used.
Further, in the case where the memory control apparatus selects the cache memory on the basis of the cache memory using state, the memory control apparatus has means for storing the using state information of each of the plurality of cache memories onto the control memory. The memory control apparatus selects which cache memory is allocated in accordance-with the using state of the cache memory. According to the above control method, by averaging the present using states of the cache memories, the cache memory space is effectively used.
On the other hand, the cache memory in the memory control apparatus of another embodiment is constructed by a plurality of unit memories which can each be closed. When a certain unit memory is closed, the cache memory including the closed unit memory is called to be in a degenerate state. When there is a cache memory in the degenerate state, a balance of the capacities among the plurality of cache memories is broken. Therefore, it is necessary to control the memory allocation in consideration of a capacity ratio of the cache memory. For this purpose, the memory control apparatus has means for holding the closed state information of each unit memory as a component element of the plurality of cache memories, calculates the capacity of each cache memory by the closed state information, and selects the cache memory to be allocated in accordance with a capacity ratio. In this instance, the above method can be also commonly used.
An embodiment of the invention will now be described hereinbelow with reference to the drawings.
Fig. 1 shows a control construction of an information processing system according to an embodiment of the invention. The information processing system has a processing unit (CPU 1), a memory device 2, and a memory control apparatus 3 connected to the processing unit 1 and memory device 2.
The memory control apparatus 3 is connected to the processing unit 1 through a channel control 31 and is connected to the memory device 2 through a drive control 32. Further, the memory control apparatus 3 includes therein: non-volatile cache memories A 33 and B 34 which are non-volatilized by a battery power source; a channel control unit 35; a drive control unit 36; a cache control unit 37; a construction control unit 38; and a control memory 39.
A construction of the memory device 2 and an outline of a flow of data will now be described hereinbelow. The memory device 2 constructs a disk array by a plurality of disk devices 21. In the disk array, data 22 is divided and divided data 221 is stored into the other different disk devices 21, respectively. Data (parity data) to guarantee the data is formed by the divided data 221 and is stored into the disk device 21 different from the disk device in which the data 221 was stored.
In the embodiment, the divided data 221 is managed on a slot unit basis. The "slot" includes a plurality of records and is a unit which is managed by a unique number called a slot number. A state in which the data corresponding to a read request and a write request issued from the processing unit 1 have already existed in the cache memory A 33 or B 34 is referred to as a read hit and a write hit, respectively. On the contrary, a state in which the requested data doesn't exist in the cache memory is referred to as a read miss and a write miss, respectively. In the data on the cache memory, the data which coincides with the data in the memory device 2 is called clean data. In the data on the cache memory, the updated data which is not stored in the memory device 2 but has been stored in only the cache memory is called dirty data.
In the case where the read request for the data 22 in the memory device 2 is issued from the processing unit 1, the data is read out by each disk control unit and, after that, the data 22 is stored into the cache memory A 33 or B 34.
In the case where the write request for the memory device 2 is issued from the processing unit 1, the data as a target to be updated is read out from the memory device 2 into the cache memory at the time of the write miss. At a time point when the updated data is written into the cache memory, the memory control apparatus 3 returns a response indicative of the end of the writing process to the processing unit (write-after control). In the writing process, the data written in the cache memory becomes dirty data. The dirty data is duplexed and stored into the cache memories A 33 and B 34. In this instance, the clean data before updating which was read out from the memory device 2 is held as it is in the cache memory.
The process to write the dirty data into the memory device 2 is executed asynchronously with the write request from the processing unit 1. New parity data is formed by the dirty data and clean data in the cache memory. The dirty data and the new parity data are written into the memory device 2. At the time point when the writing process is finished, the data as a target of the writing process doesn't need to be duplexed and held as dirty data into the cache memories. This data, accordingly, is held as clean data in either one of the two cache memories. The data can be also deleted from the cache memory.
As mentioned above, the cache memory to store the clean data and the dirty data in which the clean data was updated is called a cache surface and the cache memory to store the other one of the duplexed dirty data is called an NVS surface.
A control of the memory control apparatus which is used in the invention will now be described. The cache memory A 33 is constructed by n unit cache memories (A-1, A-2, ..., A-n) 331 and can be closed every unit cache memory 331. A state of the cache memory including the closed unit cache memory is referred to as a degenerate state. A state of the cache memory in the case where all of the unit cache memories 331 included in the cache memory is referred to as a closed state. The same shall also similarly applied to the cache memory B 34 and unit cache memories 341.
The cache control unit 37 has means 371 for allocating the cache memory to store data when the data is stored into the cache memory in association with the reading/writing process from the processing unit 1.
A construction control unit 38 manages a normal/closed state of each of the unit cache memories 331 and 341 and a normal/degenerate/closed state of each of the cache memories 33 and 34 and holds as cache construction state information 392 onto a control memory 39. The cache control unit 37 manages using states of the cache memories 33 and 34 and holds as cache using state information 391 onto the control memory 39.
Figs. 2 and 3 show the details of the cache using state information 391 and cache construction state information 392. Using state information 3911 of the cache memory A and using state information 3912 of the cache memory B are included in the cache using state information 391. Total capacities 39111 and 39121, total use amounts 39112 and 39122, dirty data amounts 39113 and 39123, and usable memory capacities 39114 and 39124 are recorded in the using state information 3911 and 3912, respectively. The total capacity is a capacity of the cache memory installed. The usable memory capacity is a normal total capacity of the unit cache memory. Cache memory A construction state information 3921 and cache memory B construction state information 3922 are included in the cache construction state information 392. Cache memory usable/unusable information 39211 and 39221 and unit cache memory usable/unusable information 39212 and 39222 are held in the construction state information 3921 and 3922, respectively. Further, the number of executing times of the selecting process of the cache memory of the cache surface, namely, the number of executing times of the allocation of the memory for the clean data is stored into the cache memory of the cache surface by the cache control unit 37. The cache control unit 37 uses those information when the cache memories are allocated.
An example of a process to select the cache memory which is used as a cache surface from the cache memory A 33 or B 34 when allocating the cache memories will now be described hereinbelow. In the embodiment, since two cache memories are provided, the selecting process of the NVS surface is unnecessary.
In the case where the read request is issued from the processing unit 1 to the memory device 2, the read data is read out from the disk device and, after that, the data 22 is stored into the cache memory selected as a cache surface by the cache control unit 37.
When the write request is issued from the processing unit 1 to the memory device 2, in case of the write miss, the cache memory which is used as a cache surface is selected by the cache control unit 37. The memory control apparatus 3 stores the updating target data which was read out from the memory device 2 into the cache memory of the cache surface and writes the updated data into the cache memory serving as a cache surface and the other cache memory serving as an NVS surface. In case of the write hit, the dirty data as updated data for the clean data is stored into the cache memory of the cache surface in which the clean data has been stored and the other cache memory as an NVS surface. In this instance, the clean data is held as it is in the cache memory of the cache surface.
An example of the process to select the cache memory that is used as a cache surface will now be described with reference to a flowchart.
Fig. 4 shows an example of the selecting process of the cache memory which is used as a cache surface. The cache control unit 37 refers to the slot number of the data which is stored into the cache memory (step 400). A check is made to see if the slot number is an even number or an odd number (step 410). When the slot number is an even number, the cache memory A 33 is selected as a cache memory which is used as a cache surface (step 420). When the slot number is an odd number, the cache memory B is selected as a cache memory which is used as a cache surface (step 430). The allocating process of the cache memories is executed (step 440). In the example, although the above judgment has been made on the basis of whether the slot number is an even number or an odd number, so long as a unit in which a deviation of data characteristics is small, it is also possible to distribute the cache surfaces every slot number and to select.
Fig. 5 shows another example of the cache memory selecting process which is used as a cache surface. The cache control unit 37 refers to the cache memory A dirty data amount 39113 in the control memory 39 and the data amount 39123 that is used as cache memory B dirty (step 500). The dirty data amounts 39113 and 39123 are compared (step 510). The cache memory of a smaller dirty data amount is selected as a cache memory which is used as a cache surface (steps 520, 530). The allocation of the cache memories is performed (step 540).
Fig. 6 shows still another example of the cache memory selecting process of the cache surface. In step 600, the cache control unit 37 refers to a counter 393. In step 610, a check is made to see if a counter value of the counter 393 is an even number or an odd number, and on the basis of a judgment result, the cache surface cache memory is selected (steps 620, 630). The cache memories are allocated in step 640.
Fig. 7 shows further another example of the cache memory selecting process of the cache surface. The cache control unit 37 refers to the cache construction state information 392 (step 700). The usable memory capacities 39114 and 39124 of the cache memories are calculated, respectively (step 710) and are recorded into the cache using state information 391. Subsequently, a usable memory capacity ratio of every cache memory surface is calculated (step 720). It is now assumed that the capacity ratio (cache memory A : cache memory B) is equal to N : M. In step 730, the slot number is referred in a manner similar to the selecting process shown in Fig. 4. A remainder in the case where the slot number is divided by (N + M) is obtained (step 740). When the remainder is equal to or less than N, the cache memory A is selected as a cache surface (step 750). In the other cases, the cache memory B is selected as a cache surface (step 760). In step 770, the cache memories are allocated. In the example, although the selection of the cache surface by the slot number has been shown, the selecting process by the counter value shown in Fig. 6 can be also used in steps 730 to 750.
The example of the selecting process of the cache memory of the cache surface has been shown above. In the embodiment, since the two cache memories are provided, the selecting process of the NVS surface is unnecessary. In case of providing N cache memories, however, the selecting process is also needed with respect to the selecting process of the NVS surface. In such a case, it will be obviously understood that the selecting process of the cache memory as a cache surface can be applied.
The memory control apparatus according to the invention can arbitrarily select the cache memory which is used as a cache surface or an NVS surface from among a plurality of cache memories. Therefore, the using states of the cache memories can be preferably balanced and there is an effect such that the using efficiency of the space of the cache memory is improved.